Re: Irix64/R10000 optimizations

Colin L. Hildinger (
Sat, 03 May 97 12:55:07 -0500

On Sat, 3 May 1997 13:14:17 -0400 (EDT), Christopher Dickens wrote:

>On Sat, 3 May 1997, Dave Ahn wrote:
>Guess what? We have someone slready working on that. The problem is, all
>UNIX clients right now are currently pure C code and we're depending on
>-O2 or -O3 on the compiler to optimize the best it can. Unfortunately
>it's not good enough. The solution? We've got a 64k bitslicing code that
>runs MUCH more efficiently on RISC architectures. According to the
>following stats we're looking at about twice the speed!
> SGI Onyx (194MHz R10000), IRIX 6.2
> My bitslice code: 1430k keys/sec
> DESCHALL irix6.2-mips client: 823k keys/sec
> DV Group's irix62-cord client: 555k keys/sec
> SolNet alpha-osf client: 589k keys/sec
> Matthew Kwan's bitslice code: 753k keys/sec
>As you can see, we're peaking here. The only thing left is machine code.
>(You don't happen to know MIPS really well do ya? 8-))
>Anyway, I've heard that this should be released publicly soon and I can't
>wait to see how much improvement I get on my R5000/SC O2.

Question - I trust that Rocke thoroughly tested his code (otherwise we
could all be wasting our time). Since the bitslice stuff is a totally
different routine, it requires full testing as well. I assume Rocke is
handling that and all the clients will be 100% reliable...

Colin L. Hildinger

Games Editor - OS/2 e-Zine!

The Ultimate OS/2 Gaming Page

The Official Unofficial AWE32 and OS/2 Warp Page